知识卡片
序列模型因果指自回归遮罩
语言模型中causal通常表示预测当前位置之后的token时只使用之前上下文。它限制计算的信息方向,不等于已学习现实DGP中的干预或反事实机制;文本生成顺序也不自动等于文本所描述事件的因果顺序。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 因果与大语言模型
原书出处 · 1 处
- · 查看原书依据
causal language models,” but these are not causal models in the way we’ve discussed in this book. They are not a causal model of a DGP. “Causal” here refers to the autoregressive nature of the model—the model evaluates the probability of a token in a sequence conditional only on the tokens that came before it.
参考来源
- 位置:《Causal AI》第13章《因果与大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-13.xhtml)
- 结论依据:摘录中关于序列模型因果指自回归遮罩的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:causal language models,” but these are not causal models in the way we’ve discussed in this book. They are not a causal model of a DGP. “Causal” here refers to the autoregressive nature of the model—the model evaluates the probability of a token in a sequence conditional only on the tokens that came before it.